Dario Amodei, Anthropic’s CEO, published a two-part policy paper on Tuesday that doesn’t pull any punches. Titled ‘Policy on the AI Exponential,’ it sketches out nothing less than a comprehensive plan for governing AI — from technical oversight to social policy.
Airplane Logic for AI Models
The core of part one, the ‘Advanced AI Framework’: frontier AI models should be treated like airplanes. Mandatory testing by independent third parties before a model can ship. And if a model doesn’t meet safety standards? The government should have the power to block or reverse its release.
That’s a remarkable position for the CEO of a company currently preparing its IPO. Amodei is essentially calling for an agency that could ban his own company’s models.
UBI Funded by AI Taxes
Part two, the ‘Economic Policy Framework,’ goes even further. Amodei acknowledges that AI could cause massive job displacement and proposes a package: wage insurance that covers the gap when someone has to take a lower-paying job. Tax credits for employers who retain workers. Training grants. And if displacement proves large and permanent — universal basic income, funded by taxes on AI companies or higher capital gains taxes.
Why Now?
According to Amodei, the trigger was Claude Mythos Preview. The model demonstrated that a frontier AI system could pose genuine threats to financial systems, critical infrastructure, and national security. That convinced him that voluntary commitments are no longer enough.
Context
You have to see this in context: Anthropic just closed its Series H round at a $965 billion valuation and filed for its IPO. A CEO who’s simultaneously preparing to go public and calling for stricter regulation is either strategically brilliant or genuinely worried. Maybe both.
The question is whether Washington listens. In a country that can’t even agree on unified AI legislation, an FAA for AI is an ambitious goal. But the fact that this proposal comes from industry — not regulators — makes it at least harder to ignore.
Sources: